CN107844921A - A kind of customer action predictor method based on embedding technologies - Google Patents

A kind of customer action predictor method based on embedding technologies Download PDF

Info

Publication number
CN107844921A
CN107844921A CN201711311722.9A CN201711311722A CN107844921A CN 107844921 A CN107844921 A CN 107844921A CN 201711311722 A CN201711311722 A CN 201711311722A CN 107844921 A CN107844921 A CN 107844921A
Authority
CN
China
Prior art keywords
user
data
embedding
client
method based
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201711311722.9A
Other languages
Chinese (zh)
Inventor
李开宇
王月超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan XW Bank Co Ltd
Original Assignee
Sichuan XW Bank Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sichuan XW Bank Co Ltd filed Critical Sichuan XW Bank Co Ltd
Priority to CN201711311722.9A priority Critical patent/CN107844921A/en
Publication of CN107844921A publication Critical patent/CN107844921A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063Operations research, analysis or management
    • G06Q10/0639Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393Score-carding, benchmarking or key performance indicator [KPI] analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • G06Q30/0202Market predictions or forecasting for commercial activities

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Accounting & Taxation (AREA)
  • Economics (AREA)
  • Finance (AREA)
  • Theoretical Computer Science (AREA)
  • Marketing (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Data Mining & Analysis (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The present invention discloses a kind of customer action predictor method based on embedding technologies, is related to internet financial information technology field, comprises the following steps:Step 1:User's history data are acquired and marked;Step 2:Build the serialized data structure between the commodity of client's purchase and user;Step 3:In serialized data structure, using embedding technologies, client's dimension coded data step 4 is trained:Monitor model training and optimization have been carried out using encoder client data;Step 5:Pass through model pre-estimating user behavior;Solve existing calculating user and state in periodic process that construction feature process is very cumbersome, can not embody big data advantage, and feature construction mode, dimension are all not perfect enough, the problem of causing model training effect to meet bottleneck.

Description

A kind of customer action predictor method based on embedding technologies
Technical field
The present invention relates to internet financial information technology field, more particularly to a kind of client based on embedding technologies Behavior predictor method.
Background technology
With the development of internet, net purchase has become daily life inalienable part.4G application in recent years With popularization, mobile electric business, social electric business obtain high speed development, and electric business coverage rate is further widened, the interaction between user and businessman Frequently.Various data of the user in electric business are also into the increase of gusher formula.The specification administered, merchandised with data;Enterprise can obtain Electric quotient data it is increasingly abundanter.Among the customer oriented market environment led, client is the most important assets of enterprise, is city The focus of field competition, numerous enterprises deploy keen competition for the limited customer resources of activity.How to look for most having enterprise The client of value is simultaneously retained, and the growth to enterprise has very important significance.Customer value is that client can be continuous Income is brought to enterprise.
Obviously, different clients is that the value that enterprise creates is different, therefore accurately defines and accurately calculate client and be full of Sharp ability is most important to enterprise.Customer development (customer lifetime value, abbreviation CLV) is exactly base It is born in this.Customer life cycle:
Customer development (customer lifetime value CLV) is to introduce to use for reference simultaneously in recent years from foreign countries Again and again the concept used.Here customer life cycle refers to client and Enterprise linkage business duration cycle, client Life cycle value then refers to the value that client brings in the contacting of the whole life cycle with bank for bank.In this sight Under the guiding of thought, we are not only it is noted that the value that client will bring for bank, also with greater need for paying attention to client in whole Life Cycle Can be that the overall value that bank brings reaches much within phase.
User behavior of estimating in this patent refers to client without buying behavior (stage) to have desire to purchase (stage) one The identification and management of individual transformation.It is exactly the process for estimating customer life cycle to estimate on the process nature of user behavior.Finally, in advance The customer action estimated is used in carry out trade marketing.
Prior art is built typically by gathering the historical data of user's the past period, using there is monitor model Mould, whether valuable commercial activity then can be produced following a period of time by the model pre-estimating user trained;And according to This implements marketing activity.This specific life cycle calculations step of method is:(1) user's history data acquisition and mark;(2) Feature Engineering processing is carried out to user's history data;It is characterized in using specialty background knowledge and skill processing data so that feature The process preferably acted on can be played on machine learning algorithm.Process contains feature extraction, feature construction, feature selecting etc. Module;(3) using has monitor model to be trained and optimize user's history data;(4) by there is monitor model to estimate user Behavior, such as LR moulds model, SVM models;Feature Engineering processing in step (2) is by original list by the historical data of user Data, statistical analysis post processing is carried out into being aggregated into categorical data.The subscriber lifecycle accuracy effect that this method obtains Fruit is dependent on " Feature Engineering processing " and " model training, optimization " the two links, and the main task of Feature Engineering is in structure Before established model training, from initial data, feature is extracted, is characterized as the input data of model training.In consideration of it, Feature Engineering Processing two aspect bottlenecks, the bottleneck that distribution is in terms of bottleneck and efficiency in terms of the effect at present.In terms of efficiency, due to Feature Engineering Processing substantially leans on people's craft construction feature at present, and construction feature process is very cumbersome, typically to pass through data analysis, statistics meter Calculate, screening and filtering these steps, it is quite time-consuming so that many data can not embody big data advantage using limited;Secondly, work Business should be familiar with, understand data analysis again by making personnel, so seldom also bad culture, can cause human cost height.Effect side Face, staff is in limitation present on individual experience and knowledge accumulation so that feature construction mode, dimension are all not complete enough Kind, model training effect meets bottleneck.
The content of the invention
It is an object of the invention to:During being estimated for the existing customer action of solution, construction feature process is very cumbersome, Typically to pass through data analysis, statistics calculates, screening and filtering these steps, quite time-consuming so that many data use limited, nothing The existing big data advantage of body of laws, staff should be familiar with business, understand data analysis again, so seldom also bad culture, can lead Cause human cost high, efficiency is low, also, staff is in limitation present on individual experience and knowledge accumulation so that special It is all not perfect enough to levy building mode, dimension, the problem of causing model training effect to meet bottleneck, the present invention provides one kind and is based on The user behavior predictor method of embedding technologies.
Technical scheme is as follows:
A kind of customer action predictor method based on embedding technologies, comprises the following steps:
Step 1:User's historical data interior for a period of time is acquired and marked;
Step 2:Serialized data between the commodity of client's purchase and user is built by the historical data of the user of collection Structure;
Step 3:In serialized data structure, using embedding technologies, client's dimension coded data is trained;
Step 4:Monitor model training and optimization have been carried out using encoder client data;
Step 5:By there is monitor model to estimate user behavior;
Specifically, the step 1 collects specifically, purchase commodity probability of the user in electric business website is entered into row set, Including user's trade name, price, quantity, time, the means of payment and number of visits.
Specifically, the step 2 concretely comprises the following steps:By the historical data of user in step 1, be built into using commodity as User's purchase sequence data structure of unit, each one commodity of behavior, while the user that will buy the commodity, press in every a line The arrangement of time buying sequencing from left to right comes out.
Further, in the step 3, user's serialized data of step 2 output is subjected to embedding models Training, obtains the embedded coding data of client, Embedding mathematically represents a maping:X->Y, that is, one Function, X here are exactly some value in data, wherein, embedding model training processes are specially:
(1) generate the vector of fixed dimension at random to each user, complete initialization;
(2) Huffman trees are built and stores the vector, its node is parameter to be optimized;
(3) stationary window is set, taking out any one has data and the enough user of data volume and several users around it, The maximum likelihood of these users and its surrounding user are calculated, to optimize the node of Huffman trees;
(4) by moving window, step (3) is performed repeatedly, completes the parameter optimization to whole Huffman trees.
Further, concretely comprising the following steps in step 4:The each encoder client data for training to obtain in step 3 are made It is characterized, input has in monitor model and is trained, and currently used model is LR (Logic Regression Models), model cross validation Accuracy rate be obviously improved.
Further, the step 5 is specially:The historical purchase data of the user estimated will be needed, carried out Embedding processing, obtains that this is subscriber-coded, then will be prejudged in coding input LR models, and anticipation score is more than 0.5 knowledge It Wei not can produce buying behavior user.
After such scheme, beneficial effects of the present invention are:
In model construction process, the work such as data statistics, analysis, selection involved by the Feature Engineering stage, it is saved;Electricity In quotient data, the present invention can be used for Embedding model trainings by what a certain commodity user time buying sequential build came out The mode of serialized data, more large-scale data can be relied on, without screening training user, solve staff individual Limitation present on the accumulation of body experience and knowledge so that all not perfect enough the bottleneck problem of feature construction mode, dimension.In spy Engineering phase is levied, by association rule mining, the mode such as assemblage characteristic search can also save the work of artificial data analysis, output Feature, but this Partial Feature easily produces over-fitting for following model, is not suitable for promoting;Model accuracy rate and ROC's Evaluation index, all there is a lifting after using new method, and specific data are as follows:
Modeling method Accuracy rate value ROC values
Traditional modeling method 79.2% 0.653
The inventive method 81.7% 0.689%
Brief description of the drawings
The user that Fig. 1 is the present invention buys data structure diagram;
Fig. 2 is embedding model training process schematics.
Embodiment
Example is applied below in conjunction with the present invention, the technical scheme in the present embodiment is clearly and completely described, is shown So, described embodiment is only the part of the embodiment of the present invention, rather than whole embodiments.Based in the present invention Embodiment, the every other embodiment that those of ordinary skill in the art are obtained under the premise of creative work is not made, all Belong to the scope of protection of the invention.
The customer action predictor method based on embedding technologies of the present embodiment, comprises the following steps:
Step 1:To user, interior historical data is acquired and marked for a period of time;Step 1 is specifically, by user in electricity Purchase commodity probability on business website enters row set and collected, including user's trade name, price, quantity, time, the means of payment and clear Look at number;Some user generates buying behavior labeled as 1 in historical data, and another part user does not produce purchase row For labeled as 0;Following model can or can not produce buying behavior according to this study anticipation user, both judge 0 or 1.
Step 2:Build the serialized data structure between the commodity of client's purchase and user;Step 2 concretely comprises the following steps: By the historical data of user in step 1, the user's purchase sequence data structure being built into units of commodity, user's purchase Data structure is bought graphically out as shown in figure 1, in Fig. 1, each one commodity of behavior, while the use that the commodity will be bought Family, come out by the arrangement of time buying sequencing from left to right in every a line, so also just produce needed for embedding technologies Serializing input data.For Embedding technologies, exactly definition object vector is showed, is the equal of to target Encode;Such as each user Zhang San, Li Si, it is very discrete data for enterprise, represents meaningless with name, and And a large amount of sparse datas can be produced.Embedding mathematically represents a mapping, f:X->Y, that is, a mapping function, Target is mapped to a vector;Mapping function is the sequence being made up of target, and structure specific structure (mainly considers target Cooccurrence relation) learnt.
Step 3:In serialized data structure, using embedding technologies, client's dimension coded data is trained;Specifically Ground, in step 3, user's serialized data of step 2 output is subjected to embedding model trainings, obtains the embedding of client Enter coded data, wherein, as shown in Fig. 2 embedding model training processes are specially:
(1) generate the vector of fixed dimension at random to each user, complete initialization;
(2) Huffman trees are built and stores the vector, its node is parameter to be optimized;
(3) stationary window is set, taking out any one has data and the enough user of data volume and several users around it, The maximum likelihood of these users and its surrounding user are calculated, to optimize the node of Huffman trees;
(4) by moving window, step (3) is performed repeatedly, completes the parameter optimization to whole Huffman trees;
Generally, the internal structure of embedding models is exactly a Huffman tree, each leaf node generation of tree One client of table, numerical value corresponding to each node is the embedded coding number of client in whole path from root node to leaf node According to coded data also arriving just after the completion of optimization.Generate the vector of fixed dimension at random to each user, that is, client Embedded coding data, are calculated by model.
Step 4:Monitor model training and optimization have been carried out using encoder client data;Concretely comprising the following steps in step 4: Using each encoder client data for training to obtain in step 3 as feature, input, which has in monitor model, to be trained, and is currently made It is LR models (Logic Regression Models) to have monitor model, and the accuracy rate of model cross validation is obviously improved;LR models (Logic Regression Models) are construed to:Generalized linear regression is to explore " expectation of response variable " and the relation of " independent variable ", with Realize certain fitting to non-linear relation.Here it is related to one " contiguous function " and one " error function ", " response becomes Linear relationship be present after contiguous function acts on, with " independent variable " in the expectation of amount ".Different " contiguous function " is chosen with " missing Difference function " can construct different generalized regressive models.When error function takes " bi-distribution " and contiguous function takes " logit letters During number ", it is exactly common " logistic regression models ", substantial amounts of application has been obtained in the problem of 0-1 is responded.Except LR moulds Type, it can also be trained using other model, such as spectrum Random Forest model, SVM models.
Step 5:By there is monitor model to estimate user behavior, such as LR, SVM model;The step 5 is specially:Need to The historical purchase data for the user to be estimated, Embedding processing is carried out, obtain that this is subscriber-coded, then by coding input LR moulds Prejudged in type, anticipation score more than 0.5 be identified as can produce buying behavior user.
The user that buying behavior is produced for the meeting finally screened carries out carry out trade marketing, and marketing type can be Different, such as short message push and App push.
It is obvious to a person skilled in the art that the invention is not restricted to the details of above-mentioned one exemplary embodiment, Er Qie In the case of without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter From the point of view of which point, embodiment all should be regarded as exemplary, and be nonrestrictive, the scope of the present invention is by appended power Profit requires rather than described above limits, it is intended that all in the implication and scope of the equivalency of claim by falling Change is included in the present invention.Any reference in claim should not be considered as to the involved claim of limitation.
Moreover, it will be appreciated that although the present specification is described in terms of embodiments, not each embodiment is only wrapped Containing an independent technical scheme, this narrating mode of specification is only that those skilled in the art should for clarity Using specification as an entirety, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art It is appreciated that other embodiment.

Claims (6)

1. a kind of customer action predictor method based on embedding technologies, it is characterised in that comprise the following steps:
Step 1:User's history data are acquired and marked;
Step 2:According to the serialized data structure between the commodity for being used for historical data structure client's purchase of collection and user;
Step 3:In serialized data structure, using embedding technologies, client's dimension coded data is trained;
Step 4:Monitor model training and optimization have been carried out using encoder client data;
Step 5:By there is monitor model to estimate user behavior.
A kind of 2. customer action predictor method based on embedding technologies according to claim 1, it is characterised in that The step 1 is specially:Purchase commodity probability of the user in electric business website is entered into row set to collect, including user's trade name, Price, quantity, time, the means of payment and number of visits.
A kind of 3. customer action predictor method based on embedding technologies according to claim 1, it is characterised in that The step 2 concretely comprises the following steps:By the historical data of user in step 1, the user's purchase being built into units of commodity Sequences data structures, each one commodity of behavior, while the user that the commodity will be bought, it is successively suitable by the time buying in every a line The arrangement of sequence from left to right comes out.
A kind of 4. customer action predictor method based on embedding technologies according to claim 1, it is characterised in that In the step 3, user's serialized data of step 2 output is subjected to embedding model trainings, obtains the insertion of client Coded data, wherein, embedding model training processes are specially:
(1) generate the vector of fixed dimension at random to each user, complete initialization;
(2) Huffman trees are built and stores the vector, its node is parameter to be optimized;
(3) stationary window is set, taking out any one there are data and the enough user of data volume and a several users around it, calculating These users and the maximum likelihood of its surrounding user, to optimize the node of Huffman trees;
(4) by moving window, step (3) is performed repeatedly, completes the parameter optimization to whole Huffman trees.
A kind of 5. customer action predictor method based on embedding technologies according to claim 1, it is characterised in that Concretely comprising the following steps in step 4:The each encoder client data for training to obtain in step 3 there is into supervision as feature, input It is trained in model,.
A kind of 6. customer action predictor method based on embedding technologies according to claim 1, it is characterised in that The step 5 is specially:The historical purchase data of the user estimated will be needed, carry out Embedding processing, obtain the user Coding, then will be prejudged in coding input LR models, and anticipation score being identified as more than 0.5 can produce buying behavior user.
CN201711311722.9A 2017-12-11 2017-12-11 A kind of customer action predictor method based on embedding technologies Pending CN107844921A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711311722.9A CN107844921A (en) 2017-12-11 2017-12-11 A kind of customer action predictor method based on embedding technologies

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711311722.9A CN107844921A (en) 2017-12-11 2017-12-11 A kind of customer action predictor method based on embedding technologies

Publications (1)

Publication Number Publication Date
CN107844921A true CN107844921A (en) 2018-03-27

Family

ID=61663393

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711311722.9A Pending CN107844921A (en) 2017-12-11 2017-12-11 A kind of customer action predictor method based on embedding technologies

Country Status (1)

Country Link
CN (1) CN107844921A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110210913A (en) * 2019-06-14 2019-09-06 重庆邮电大学 A kind of businessman frequent customer's prediction technique based on big data

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140376804A1 (en) * 2013-06-21 2014-12-25 Xerox Corporation Label-embedding view of attribute-based recognition
US9037464B1 (en) * 2013-01-15 2015-05-19 Google Inc. Computing numeric representations of words in a high-dimensional space
US20160188724A1 (en) * 2014-12-24 2016-06-30 Facebook, Inc. Tag prediction for content based on user metadata
US9536248B2 (en) * 2006-02-22 2017-01-03 24/7 Customer, Inc. Apparatus and method for predicting customer behavior
US20170140280A1 (en) * 2010-09-23 2017-05-18 24/7 Customer, Inc. Predictive customer service environment

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9536248B2 (en) * 2006-02-22 2017-01-03 24/7 Customer, Inc. Apparatus and method for predicting customer behavior
US20170140280A1 (en) * 2010-09-23 2017-05-18 24/7 Customer, Inc. Predictive customer service environment
US9037464B1 (en) * 2013-01-15 2015-05-19 Google Inc. Computing numeric representations of words in a high-dimensional space
US20140376804A1 (en) * 2013-06-21 2014-12-25 Xerox Corporation Label-embedding view of attribute-based recognition
US20160188724A1 (en) * 2014-12-24 2016-06-30 Facebook, Inc. Tag prediction for content based on user metadata

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
ALEXANDRE PASSOS等: ""Lexicon infused phrase embeddings for named entity resolution"", 《ARXIV PREPRINT ARXIV:1404.5367》 *
E.SHIJIA等: "《China Conference on Knowledge Graph and Semantic Computing》", 23 November 2016 *
MARCO ANTONIO SOTO HERNANDEZ等: "《2013 International Conference on Reconfigurable Computing and FPGAs (ReConFig)》", 6 February 2014 *
SANJAY BAJPAI等: ""Enhancement of Security and Embedding Capacity through Huffman Coding in Steganography"", 《INTERNATIONAL JOURNAL OF EMERGING TRENDS & TECHNOLOGY IN COMPUTER SCIENCE(IJETTCS)》 *
裴楠等: ""基于计数模型的Word Embedding算法"", 《沈阳航空航天大学学报》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110210913A (en) * 2019-06-14 2019-09-06 重庆邮电大学 A kind of businessman frequent customer's prediction technique based on big data

Similar Documents

Publication Publication Date Title
CN108154430A (en) A kind of credit scoring construction method based on machine learning and big data technology
Wong et al. A hybrid intelligent model for medium-term sales forecasting in fashion retail supply chains using extreme learning machine and harmony search algorithm
CN106326248B (en) The storage method and device of database data
Zhao et al. Sales forecast in e-commerce using convolutional neural network
CN103514566A (en) Risk control system and method
CN108536536A (en) Resource consolidation and optimization system and method under a kind of cloud manufacturing environment based on quantum genetic algorithm
CN102663616A (en) Method and system for measuring web advertising effectiveness based on multiple-contact attribution model
CN110147389B (en) Account processing method and device, storage medium and electronic device
CN108388955A (en) Customer service strategies formulating method, device based on random forest and logistic regression
CN113469730A (en) Customer repurchase prediction method and device based on RF-LightGBM fusion model under non-contract scene
CN109492104A (en) Training method, classification method, system, equipment and the medium of intent classifier model
CN108133390A (en) For predicting the method and apparatus of user behavior and computing device
Krejci et al. Modeling food supply chains using multi-agent simulation
CN116245399A (en) Model training method and device, nonvolatile storage medium and electronic equipment
CN117291655B (en) Consumer life cycle operation analysis method based on entity and network collaborative mapping
CN110009427B (en) Intelligent electric power sale amount prediction method based on deep circulation neural network
Lin et al. Optimization analysis for an infinite capacity queueing system with multiple queue-dependent servers: genetic algorithm
Yang Enterprise Network Marketing Prediction Using the Optimized GA‐BP Neural Network
CN107239853B (en) Intelligent housekeeper system based on cloud computing and working method thereof
CN107844921A (en) A kind of customer action predictor method based on embedding technologies
Venkatesan et al. [Retracted] A Crop Growth Prediction Model Using Energy Data Based on Machine Learning in Smart Farms
CN107992558A (en) A kind of electricity consumption user clustering method based on the improved K-means algorithms of genetic algorithm
CN116883057A (en) XGBoost-based high-precision power customer marketing channel preference prediction system
CN105023041B (en) Data clustering method and B2B platform customer priorities acquisition methods, device
CN116542490A (en) Virtual power plant day-ahead dispatching encapsulation model and dispatching ex-definition model construction method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20180327

RJ01 Rejection of invention patent application after publication